Search

Accurate short-term load forecasting (STLF) is of positive significance to the effective management of power companies and the stable operation of society. In spite of many studies conducted in this field, there are few to consider the inherent disadvantages of an individual module, which results in sub-optimal forecasting accuracy. Therefore, by integrating data preprocessing module and optimization module into support vector regression (SVR) forecasting module, this paper successfully presents a novel model (AHO-Lasso-SVR). The data preprocessing module, which is comprised of feature construction and Lasso feature selection, is used to construct and select meaningful features. An adaptive hybrid optimization (AHO) algorithm is proposed by introducing two strategies on the basis of standard particle swarm optimization (PSO). The AHO algorithm inevitably increases the computational complexity of model learning, thus, this paper proposes a subsampling technology to improve the optimization efficiency of the algorithm based on the sparsity of SVR. The proposed model is used to forecast the load at 48 points in the next day. To verify the properties of the proposed model, power load data from New South Wales, Australia are adopted as a case study. The results reveal that our model positively exceeds all comparison models in terms of forecasting accuracy and stability. [ABSTRACT FROM AUTHOR]

Short-term power load forecasting is an important factor affecting the security and economic operation of modern power systems. In order to avoid the limitation of a single model, this paper proposes a novel variable weight combined model that combines three single models of support vector regression (SVR), linear regression (LR) and random forest (RF). In the proposed model, a weight table is created based on time similarity, which contains weight information of 336 timestamps for determining the weight of each model. Then, particle swarm optimization (PSO) is employed to complete the modeling of the variable weight combined model. To evaluate the forecasting ability of the combined model, this paper takes the half-hourly power data of New South Wales as an example. Experimental analysis shows that all the evaluation metrics of the proposed model are better than the three single models and combined model based on average weight. [ABSTRACT FROM AUTHOR]